Ask The Expert:QoS on Catalyst Switches - Page 4

ciscomoderator · ‎08-12-2011

With Shashank Singh

Welcome to the Cisco Support Community Ask the Expert conversation. This is an opportunity to get an update on QoS on Catalyst 2960, 3550, 3560, 3750, 4500 and 6500 series switches with Cisco expert Shashank Singh. Shashank graduated in 2009 with a bachelor's degree in Computer Science and Engineering from VIT University, Vellore India. Prior to joining Cisco he worked at General Electric as a software engineer. Later on he joined the Cisco Technical Assistance Center as an engineer in October of 2009. He has been working on LAN Switching technologies in TAC since then. Shashank also holds a CCNP certificate. QoS on Catalyst switches is one of the areas of his interest.

Remember to use the rating system to let Shashank know if you have received an adequate response.

Shashank might not be able to answer each question due to the volume expected during this event. Remember that you can continue the conversation on the Network InfrastructureLAN Switching discussion forum shortly after the event. This event lasts through August 26 , 2011. Visit this forum often to view responses to your questions and the questions of other community members.

Shashank Singh · ‎08-19-2011

Hi Jon,

To be honest, I have not come across anything that explains how packet size could be a parameter for deciding traffic that needs priority. As far as I know, priority is a relative concept and if bandwidth is present in abundance, there will not be a need to prioritize anything.I agree with the thought that when there is no congestion, there is no priority bandwidth. Its just bandwidth

Cheers,

Shashank

Joseph W. Doherty · ‎08-19-2011

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Jon, you might be thinking of the need to priorize VoIP traffic by fragmenting larger packets on low bandwidth links.

Jon Marshall · ‎08-20-2011

Joseph

You may well be right because i can't find it anywhere else

Jon

jgauthray · ‎08-19-2011

Dear Shashank,

I got a customer who has L2 WAN links (SDSL 2Mbps & 4Mbps) that connect his branch offices to his central office. As there is no QoS on those links we have to limit the bandwidth on egress, and as he asked for a layer 2 WAN architecture we plan to deploy switch like C3560 or C2960 (cheaper).

I wonder if limit the bandwidth with L2 switch like C2960 is a good idea. It seems that it's possible to limit it with "srr-queue bandwidth " command but I wonder if it's the best solution :

I'm afraid of the performance impact and high CPU processing
Is implementing policy-map a best solution ?
Can we implement WRED drop or shaping policy on 2960 ?

Thanks a lot

Shashank Singh · ‎08-20-2011

Hi Julien,

srr-queue bandwidth limit lets you limit the output on a port and can be used if this serves your purpose. Policy map can be used to police traffic but this is supported only on the ingress.

Shaping can be configured only for the egress queues. None of these features are known to contribute towards persistent increase in CPU utilization. However, if your links on the switch gets congested there may be some performance issues if certain traffic gets lesser resources than required. This would requires the configuration to be fine tuned to suit the traffic.

Cheers,

Shashank

ewood2624 · ‎08-19-2011

What's the best practice for QoS trunk links trust cos or trust dscp? Also is there a best practice guide for QoS on wireless networks?

Shashank Singh · ‎08-19-2011

Hi,

It depends on how your source is marking the traffic. If source is marking dscp, trust dscp and likewise for cos. If source is marking both, you have an option. Switches always find an equivalent dscp irrespctive of the trust state configured. This internal dscp is used for applying QoS to the traffic. I would personally go with dscp as it keeps confusions away and makes things a little easier to troubleshoot.

The only best practice is to trust either cos or dscp throughout your network. Avoid trusting cos on some interfaces and dscp on others.

Cheers,

Shashank

Jonathan Hall · ‎08-19-2011

Hi Shashank,

jon.marshall's questions above have reminded me of a confusion I've always had with the egress queue SRR on 3560/3750 switches.

I understand if a port has its default SRR settings in place of shaped queue weights 25 0 0 0 and shared queue weights 25 25 25 25 that Q1 will get upto, but no more than, 1/25 (4%) of the bandwidth and Q2, Q3 and Q4 will all get, at a minimum, an equal share of the remaining 24/25 of the bandwidth, 8/25 (32%) each.

Is the shaped Q1 serviced by the scheduler at the same rate as the shared queues, or is it given a higher priority until it reaches its bandwidth limit?

If Q1 is configured as priority out the documentation says that the shaped and shared weightings are ignored and Q1 is serviced until it is empty. In this situation is Q1 not bandwidth limited in any way and could it mean if Q1 is sufficently busy packets in Q2, Q3 and Q4 never get sent?

Thanks,

Jonathan

Shashank Singh · ‎08-20-2011

Hi Jonathan,

The amount of bandwidth available with a queue at any point of time determines the rate at which scheduler will empty the queue. Shaping gurantees a constant bandwidth while sharing guarantees a minimum bandwidth. There is no priority involved with these concepts. If a shared queue has more bandwidth than a shaped queue, scheduler will take out more packets from the shared queue.

You are correct. Priority queue is serviced until empty and may use all the bandwidth available if need be. This also means that the priority queue is emptied as soon as possible. As long as there are packets in the priority queue, no other queue is serviced.

Cheers,

Shashank

leeflight · ‎08-23-2011

Hi,

I have been using the Cisco Smart Business Architecture guides

http://www.cisco.com/en/US/docs/solutions/Enterprise/Borderless_Networks/Smart_Business_Architecture/SBA_Enterprise_BN_LAN_Configuration_Guide_H2CY10.pdf

looking at access ports it shows e.g. on page 7:

!

interface GigabitEthernet1/0/2

switchport access vlan 180

switchport mode access

switchport voice vlan 181

srr-queue bandwidth share 1 30 35 5

queue-set 2

priority-queue out

mls qos trust device cisco-phone

mls qos trust cos

macro description AccessEdgeQoS

auto qos voip cisco-phone

!

I cannot find anything in the auto qos configuration that would have generated the

queue-set 2

configuration and indeed looking at the QoS model I cannot see why it would not be queue-set 1 (the default).

Any idea where the "queue-set 2" comes from?

Thanks

Lee Flight

Shashank Singh · ‎08-23-2011

Hi Lee,

queue-set 2 is generated by the auto qos command. Sometimes, queue buffers and thresholds need to be altered in order to suit traffic for which auto qos is being used. On 2900/3700 switches, every interface by default uses queue-set 1. Hence if you change queue-set 1, the changes get reflected to all the interfaces.

Keeping this in mind, auto qos makes the port use queue-set 2 (which by default has same configuration as queue set 1), so that any changes made to queue-set 2 are just reflected for the interfaces using queue-set 2.

If you want the interface using auto qos to use queue-set 1, you can explictly configure "no queue-set 2" under that interface You can review both queue-set configuration using "sh mls qos queue-set" command.

Cheers,

Shashank

leeflight · ‎08-24-2011

Hi Shashank,

thanks for your reply; I was expecting that "queue-set 2" when I read the guide however when I apply

auto qos voip cisco-phone

on a port a 2960-S as in the guide (or even on a 3750-X) I do not see "queue-set 2" in the resulting configuration

when running IOS 12.2 (latest). If I run "debug auto qos" before enabling auto qos on the interface the debug output contains:

no queue-set 1

queue-set 1

this seems to differ from the guide. So does this actually come from auto qos or does the queue-set configuration need to be separately applied? We would like to follow the SBA best practice but we are not clear on this point and cannot find anything in the guide that clarifies this.

Thanks

Lee Flight

Shashank Singh · ‎08-24-2011

Hi Lee,

I tested this out on WS-C2960S-48TS-S running 12.2(53)SE and found that queue-set 2 is applied when I configure auto qos on the interface.

*Mar 19 18:15:34.051 JST: mls qos trust cos

*Mar 19 18:15:34.591 JST: no queue-set 1

*Mar 19 18:15:34.602 JST: queue-set 2

*Mar 19 18:15:34.607 JST: priority-queue out

*Mar 19 18:15:34.612 JST: srr-queue bandwidth share 1 30 35 5

However, queue-set 2 is NOT applied on a 3750 interface on application of auto-qos. This means that the generated configuration is different between different platforms.

Cheers,

Shashank

leeflight · ‎08-25-2011

Hi Shashank,

thanks for doing the test. What I have is

WS-C2960S-48FPS-L running 12.2(58)SE1 and I get

!
interface GigabitEthernet1/0/3

srr-queue bandwidth share 1 30 35 5
priority-queue out
mls qos trust dscp
auto qos trust dscp
spanning-tree portfast
!

a the resultant interface configuration when using "auto qos trust dscp".

I also tried on 15.0.(1)SE and get the same. I have not tried a lower IOS
revision.

Can I ask which auto qos option you used on the interface in your test?

Thanks for your help with this,
Lee Flight

Shashank Singh · ‎08-25-2011

Hi Lee,

I tried auto qos voip trust.

Cheers,

Shashank